Neural representations of honesty predict future trust behavior

Abstract

Theoretical accounts propose honesty as a central determinant of trustworthiness impressions and trusting behavior. However, behavioral and neural evidence on the relationships between honesty and trust is missing. Here, combining a novel paradigm that successfully induces trustworthiness impressions with functional MRI and multivariate analyses, we demonstrate that honesty-based trustworthiness is represented in the posterior cingulate cortex, dorsolateral prefrontal cortex and intraparietal sulcus. Crucially, brain signals in these regions predict individual trust in a subsequent social interaction with the same partner. Honesty recruited the ventromedial prefrontal cortex (VMPFC), and stronger functional connectivity between the VMPFC and temporoparietal junction during honesty encoding was associated with higher trust in the subsequent interaction. These results suggest that honesty signals in the VMPFC are integrated into trustworthiness beliefs to inform present and future social behaviors. These findings improve our understanding of the neural representations of an individual's social character that guide behaviors during interpersonal interactions.

Fig. 1

Paradigms. a Schematic representation of the take advice game (TAG). Advisers were given information about one of the two cards and could communicate this information to the advisee. Participants, in the role of advisee, made a decision based on the information received (decision phase). In the feedback phase, advisees saw the actual numbers on the cards, which informed them about the adviser’s honest behavior (honest vs. dishonest), and a green or red circle, which informed them whether they won or lost, respectively. b After the TAG, participants in the role of investor played a one-shot trust game (TG) with the advisers now in the role of trustee. Investors received a monetary endowment and decided whether they wanted to entrust some of this amount with the trustees. Investors were told that the shared amount was tripled by the experimenter and passed on to the trustee, who could decide to share back any portion of the tripled amount. See also Supplementary Fig. 1

Fig. 2

Behavioral results. a Trusting behavior in the take advice game over runs (left) and on average (right) toward honest and dishonest advisers. On average, participants took significantly more advice from the honest than the dishonest adviser (t test). Data points on the left were interpolated for visualization purposes and shadowed areas represent standard errors. White lines in the box-plots on the right represent average advice-taking behavior across participants. Each black dot represents one participant. b Amount of money entrusted in the trust game with honest (left) and dishonest (right) others correlated with participants’ willingness to take advice from the advisers (Spearman’s correlations). Each dot represents one participant. ***P < 0.001

Fig. 3

Decoding honesty and predicting trust. In two MVPAs applied to the feedback phase of the TAG (a), a support vector machine (SVM) was trained to decode honest and dishonest advice (GLM1) to determine the trustworthiness decoding network (upper), and to decode winnings and losses (GLM2) to determine the value decoding network (lower). The trustworthiness decoding network (b) included brain regions such as the PCC, DLPFC, and IPS, and could successfully distinguish neural signatures of honesty and dishonesty in out-of-sample individuals (c). The value decoding network (d) included the striatum and ACC and could successfully distinguish neural signatures of winnings and losses in out-of-sample individuals (e). Finally, a multivariate prediction analysis with support vector regression (SVR) showed that the neural patterns of the trustworthiness decoding network successfully predicted individual economic trust decisions in the TG, thereby showing across-context generalizability (f). Both out-of-sample classification and prediction analyses were based on a leave-one-subject-out cross-validation procedure and their significance tested using a permutation test with 10,000 permutations. Each dot represents one participant. See also Supplementary Fig. 2 and Supplementary Table 1. MVPA, multivariate voxel pattern analysis; TAG, take advice game; PCC, posterior cingulate cortex; IPS, intraparietal sulcus; DLPFC, dorsolateral prefrontal cortex; ACC, anterior cingulate cortex; TG, trust game. Heatmap represents t values

Fig. 4

Honesty vs. Dishonesty. Univariate contrasts revealed that brain areas within the trustworthiness decoding network (i.e., IPL and DLPFC) were more engaged by dishonesty than honesty (a), whereas honesty more strongly recruited the VMPFC (b). Error bars indicate standard errors across participants. Each dot represents one participant. See also Supplementary Table 2. IPL, inferior parietal lobule; DLPFC, dorsolateral prefrontal cortex; VMPFC, ventromedial prefrontal cortex; a.u., arbitary units. Heatmap represents t values

Fig. 5

Task-based functional connectivity analysis. Task-based functional connectivity between the VMPFC and left pTPJ was stronger for honesty than dishonesty (a). Critically, this functional connectivity correlated with an individual’s willingness to trust in the TG (b), but not with one’s payoffs in the TAG (c) (Spearman’s correlations). Blue dots on correlation plots on the left represent behaviors toward honest advisers, orange dots on correlation plots on the right represent behaviors toward dishonest advisers. Each dot represents one participant. VMPFC, ventromedial prefrontal cortex; pTPJ, posterior temporo-parietal junction; PPI, psychophysiological interaction; a.u., arbitary units. Heatmap represents t values

Fig. 6

Modulation of neural responses to value information. Whole-brain contrast analyses from GLM3 on the feedback phase yielded significant activations in the parietal cortex for dishonesty during both positive (a) and negative outcomes (b). Honesty, on the contrary, modulated only positive outcomes in the OFC (c). An ROI analysis (d) indicated higher activity in the OFC in response to positive outcomes when interacting with honest advisers (t test). Error bars indicate standard errors across participants. Each dot represents one participant. See also Supplementary Tables 3–5. ***P < 0.001. ROI, region of interest; IPL, inferior parietal lobule; IPS, intraparietal sulcus; OFC, orbitofrontal cortex; a.u., arbitrary units. Heatmap represents t values